03-10-2021 06:26 AM - last edited on 03-05-2024 01:58 AM by ROGBot
04-21-2023 10:20 PM
Sure, you can do 4 M.2 drives on the Hyper M.2 card in PCIEX16_1, but you then must leave PCIEX16_2 empty. Your GPU would have to go in PCIEX16_3, running via the chipset with just x4 bandwidth that is shared with all other devices on the chipset. That's quite likely to limit the performance of a 3080 Ti in many workloads. If you are ok with a bandwidth-limited GPU, you can have a x16 storage solution.
None of the desktop platforms are really suitable for running with both high GPU performance and high PCIe storage performance at the same time. They are designed to give you either x16 + x4, or x8 + x8 + x4, more or less, in terms of CPU PCIe lanes (AM5 adds an extra x4 to that; while LGA1700 adds those to the chipset link). You can have high GPU, low storage; or high storage, low GPU. That's for all of AM4 & AM5 and Intel LGA1200 & LGA1700. You need to go up to Threadripper or Xeon workstation platforms to get the PCIe lanes (including the now quite old Intel Core-X X299 HEDT platform, which was Xeon sold as Core), or EPYC or Xeon server platforms.
I don't know your specific application, but I suggest considering something like an ASUSTOR NAS connected by 10G Ethernet for the capacity, and a smaller amount of NVMe storage for your active datasets. That's if you need to find a way to get high GPU performance, and high capacity storage. Also, don't ignore those SATA connectors, a RAID 0 or RAID 10 of 4 SATA SSDs can still deliver some useful local storage that's fast but not quite as fast as NVMe.